Software Design Lecture 11

Content Design Patterns Creational Patterns Structural Patterns Behavioral Patterns Observer Strategy State Command Chain of Responsibility

References Erich Gamma, et.al, Design Patterns: Elements of Reusable Object- Oriented Software, Addison Wesley, 1994, ISBN 0-201-63361-2. Univ. of Timisoara Course materials

Behavioral patterns are concerned with algorithms and the assignment of responsibilities between objects. describe patterns of communication between classes/objects. class patterns use inheritance to distribute behavior between classes. object patterns use object composition rather than inheritance.

Observer Pattern

Weather Monitoring Application Station Humidity Sensor Temp Pressure Weather Data Object Display Device Pulls Data displays

What needs to be done? WeatherData getTemperature() getHumidity() getPressure() measurementsChanged() /* Call this method whenever measurements are updated */ public void measurementsChanged() { // your code goes here } Update three different displays

Problem specification WeatherData class has three getter methods measurementsChanged() method called whenever there is a change Three display methods needs to be supported: current conditions, weather statistics and simple forecast System should be expandable

First cut at implementation public class WeatherData { public void measurementsChanged(){ float temp = getTemperature(); float humidity = getHumidity(); float pressure = getPressure(); currentConditionsDisplay.update (temp, humidity, pressure); statisticsDisplay.update (temp, humidity, pressure); forecastDisplay.update (temp, humidity, pressure); } // other methods

First cut at implementation public class WeatherData { public void measurementsChanged(){ float temp = getTemperature(); float humidity = getHumidity(); float pressure = getPressure(); currentConditionsDisplay.update (temp, humidity, pressure); statisticsDisplay.update (temp, humidity, pressure); forecastDisplay.update (temp, humidity, pressure); } // other methods Area of change which can be Managed better by encapsulation By coding to concrete implementations there is no way to add additional display elements without making code change

Basis for observer pattern Fashioned after the publish/subscribe model Works off similar to any subscription model Buying newspaper Magazines List servers The Observer Pattern defines a one-to-many dependency between objects so that when one object changes state, all of its dependents are notified and updated automatically.

Observer Pattern – Class diagram <<interface>> Subject registerObserver() removeObserver() notifyObservers() Observer Update() observers ConcreteSubject ConcreteObserver subject

Observer pattern – power of loose coupling The only thing that the subject knows about an observer is that it implements an interface Observers can be added at any time and subject need not be modified to add observers Subjects and observers can be reused or modified without impacting the other [as long as they honor the interface commitments]

Observer Pattern – Weather data <<interface>> Subject registerObserver() removeObserver() notifyObservers() Observer update() observers WeatherData getTemperature() getPressure() measurementsChanged() subject DisplayElement display() StatisticsDisplay ForecastDisplay CurrentConditionsDisplay

Weather data interfaces public interface Subject { public void registerObserver(Observer o); public void removeObserver(Observer o); public void notifyObservers(); } public interface Observer { public void update(float temp, float humidity, float pressure); public interface DisplayElement { public void display();

Implementing subject interface public class WeatherData implements Subject { private ArrayList observers; private float temperature; private float humidity; private float pressure; public WeatherData() { observers = new ArrayList(); } …}

Register and unregister public void registerObserver(Observer o) { observers.add(o); } public void removeObserver(Observer o) { int i = observers.indexOf(o); if (i >= 0) { observers.remove(i);

Notify methods public void notifyObservers() { for (int i=0; i<observers.size(); i++) { Observer observer = (Observer)observers.get(i); observer.update(temperature, humidity, pressure); } public void measurementsChanged() { notifyObservers()

Observer pattern More analysis

Push or pull The notification approach used so far pushes all the state to all the observers One can also just send a notification that some thing has changed and let the observers pull the state information Java observer pattern support has built in support for both push and pull in notification java.util.Observable java.util.Observer

Java Observer Pattern – Weather data Observable addObserver() deleteObserver() notifyObservers() setChanged() <<interface>> Observer update() observers WeatherData getTemperature() getPressure() measurementsChanged() subject DisplayElement display() StatisticsDisplay ForecastDisplay CurrentConditionsDisplay Observable is a class And not an interface

Problems with Java implementation Observable is a class You have to subclass it You cannot add observable behavior to an existing class that already extends another superclass You have to program to an implementation – not interface Observable protects crucial methods Methods such as setChanged() are protected and not accessible unless one subclasses Observable. You cannot favor composition over inheritance. You may have to write your own observer interface if Java utilities don’t work for your application

Changing the "Guts" of an Object ... Control "shield" the implementation from direct access (Proxy) Decouple let abstraction and implementation vary independently (Bridge) Optimize use an alternative algorithm to implement behavior (Strategy) Alter change behavior when object's state changes (State)

Strategy Pattern

Java Layout Managers GUI container classes in Java frames, dialogs, applets (top-level) panels (intermediate) Each container class has a layout manager determine the size and position of components 20 types of layouts ~40 container-types imagine to combine them freely by inheritance Consider also sorting... open-ended number of sorting criteria

Basic Aspects Intent Define a family of algorithms, encapsulate each one, and make them interchangeable Let the algorithm vary independently from clients that use it Applicability You need different variants of an algorithm An algorithm uses data that clients shouldn't know about avoid exposing complex, algorithm-specific data structures Many related classes differ only in their behavior configure a class with a particular behavior

Structure

Participants Strategy declares an interface common to all supported algorithms. Context uses this interface to call the algorithm defined by a ConcreteStrategy ConcreteStrategy implements the algorithm using the Strategy interface Context configured with a ConcreteStrategy object may define an interface that lets Strategy objects to access its data

Consequences Families of related algorithms usually provide different implementations of the same behavior choice decided by time vs. space trade-offs Alternative to subclassing see examples with layout managers We still subclass the strategies… Eliminates conditional statements many conditional statements  "invitation" to apply Strategy!

Issues Who chooses the strategy? Client Context

Implementation How does data flow between Context and Strategies? Approach 1: take data to the strategy decoupled, but might be inefficient Approach 2: pass Context itself and let strategies take data Context must provide a more comprehensive access to its data => more coupled In Java, the strategy hierarchy might be inner classes Making Strategy object optional provide Context with default behavior if default used no need to create Strategy object don't have to deal with Strategy unless you don't like the default behavior

Decorator vs. Strategy

State Pattern

Example: SPOP SPOP = Simple Post Office Protocol used to download emails from server SPOP supports the following commands: USER <username> PASS <password> LIST RETR <message number> QUIT USER & PASS commands USER with a username must come first PASS with a password or QUIT must come after USER If the username and password are valid, the user can use other commands

SPOP (contd.) LIST command Arguments: a message-number (optional) Returns: size of message in octets if message number, returns the size of that message otherwise return size of all mail messages in the mail-box RETR command Arguments: a message number Returns: the mail message indicated by that number QUIT command Arguments: none updates mailbox to reflect transactions taken during the transaction state, the logs user out if session ends by any method except the QUIT command, the updates are not done

SPOP States

The "Dear, Old" Switches in Action Think about adding a new state to the protocol... Why? object's behavior depends on its state

Basic Aspects of State Pattern Intent allow an object to alter its behavior when its internal state changes object will appear to change its class Applicability object's behavior depends on its state it must change behavior at run-time depending on that state operations with multipart conditional statements depending on the object's state state represented by one or more enumerated constants several operations with the same (or similar) conditional structure

Structure

Participants Context defines the interface of interest for clients maintains an instance of ConcreteState subclass State defines an interface for encapsulating the behavior associated with a particular state of the Context ConcreteState each subclass implements a behavior associated with a state of the Context

Collaborations Context delegates state-specific requests to the State objects the Context may pass itself to the State object if the State needs to access it in order to accomplish the request State transitions are managed either by Context or by State see discussion on the coming slides Clients interact exclusively with Context but they might configure contexts with states e.g initial state

Consequences Localizes state-specific behavior and partitions behavior for different states Put all behavior associated with a state in a state-object Easy to add new states and transitions context becomes O-C Behavior spread among several State subclasses number of classes increases, less compact than a single class good if many states... Makes state transitions explicit Not only a change of an internal value States receive a full-object status! Protects Context from inconsistent internal states

Applying State to SPOP class SPop { private SPopState state = new Start(); public void user( String userName ) { state = state.user( userName ); } public void pass( String password ) { state = state.pass( password ); public void list( int messageNumber ) { state = state.list( massageNumber ); // . . . }

SPOP States

How much State in the State? Let's identify the roles... SPop is the Context SPopState is the abstract State Start, HaveUserName are ConcreteStates All the state and real behavior is in SPopState and subclasses this is an extreme example In general Context has data and methods besides State & State methods this data will not change states Only some aspects of the Context will alter its behavior

Who Defines the State transition? The Context if ... ...states will be reused in different state machines with different transitions ... the criteria for changing states are fixed

OR…The States More flexible to let State subclasses specify the next state

State versus Strategy Rate of Change Strategy Context object usually contains one of several possible ConcreteStrategy objects State Context object often changes its ConcreteState object over its lifetime Visibility of Change All ConcreteStrategy do the same thing, but differently Clients do not see any difference in behavior in the Context ConcreteState acts differently Clients see different behavior in the Context

Command Pattern

Menu Items Use Commands

Basic Aspects Intent Encapsulate requests as objects, letting you to: parameterize clients with different requests queue or log requests support undoable operations Applicability Parameterize objects Specify, queue, and execute requests at different times Support undo recover from crashes  needs undo operations in interface Support for logging changes recover from crashes  needs load/store operations in interface Model transactions

Structure Transforms: concreteReceiver.action() in command.execute() create set Holds command Transforms: concreteReceiver.action() in command.execute()

Participants Command declares the interface for executing the operation ConcreteCommand binds a request with a concrete action Invoker asks the command to carry out the request Receiver knows how to perform the operations associated with carrying out a request. Client creates a ConcreteCommand and sets its receiver

Collaborations Client  ConcreteCommand creates and specifies receiver Invoker  ConcreteCommand ConcreteCommand  Receiver

Consequences Decouples Invoker from Receiver Commands are first-class objects can be manipulated and extended Composite Commands see also Composite pattern Easy to add new commands Invoker does not change it is Open-Closed Potential for an excessive number of command classes

Example: Open Document

Intelligence of Command objects "Dumb" delegate everything to Receiver used just to decouple Sender from Receiver "Genius" does everything itself without delegating at all useful if no receiver exists let ConcreteCommand be independent of further classes "Smart" find receiver dynamically

Undoable Commands Need to store additional state to reverse execution receiver object parameters of the operation performed on receiver original values in receiver that may change due to request receiver must provide operations that makes possible for command object to return it to its prior state History list Sequence of commands that have been executed used as LIFO with reverse-execution  undo used as FIFO with execution  redo Commands may need to be copied when state of commands change by execution

Composed Commands

Chain of Responsibility Pattern

Basic Aspects Intent Decouple sender of request from its receiver by giving more than one object a chance to handle the request Put receivers in a chain and pass the request along the chain until an object handles it Motivation context-sensitive help a help request is handled by one of several UI objects Which one? depends on the context The object that initiates the request does not know the object that will eventually provide the help

When to Use? Applicability more than one object may handle a request and handler isn't known a priori set of objects that can handle the request should be dynamically specifiable send a request to several objects without specifying the receiver

Structure

Participants & Collaborations Handler defines the interface for handling requests may implement the successor link ConcreteHandler either handles the request it is responsible for ... ... or it forwards the request to its successor Client initiates the request to a ConcreteHandler object in the chain

The Context-Help System

Consequences Reduced Coupling frees the client (sender) from knowing who will handle its request sender and receiver don't know each other instead of sender knowing all potential receivers, just keep a single reference to next handler in chain. simplify object interconnections Flexibility in assigning responsibilities to objects responsibilities can be added or changed chain can be modified at run-time Requests can go unhandled chain may be configured improperly!!

How to Design Chains of Commands? Like the military a request is made it goes up the chain of command until someone has the authority to answer the request

Implementing the Successor Chain Define new link Give each handler a link to its successor Use existing links concrete handlers may already have pointers to their successors references in a part-whole hierarchy can define a part's successor spares work and space ... ... but it must reflect the chain of responsibilities that is needed

Representing Multiple Requests using One Chain Each request is hard-coded convenient and safe not flexible (limited to the fixed set of requests defined by handler)

Unique handler with parameters more flexible but it requires conditional statements for dispatching request less type-safe to pass parameters

Decorator vs. Chain of Responsibility Comparable to “event-oriented” architecture Comparable to layered architecture (layers of an onion) The "filter" objects are of equal rank A "core" object is assumed, all "layer" objects are optional User views the chain as a "launch and leave" pipeline User views the decorated object as an enhanced object A request is routinely forwarded until a single filter object handles it. many (or all) filter objects could contrib. to each request's handling.  A layer object always performs pre or post processing as the request is delegated. All the handlers are peers (like nodes in a linked list) – "end of list" condition handling is required. All the layer objects ultimately delegate to a single core object - "end of list" condition handling is not required.   

Next time Quality attributes Representation Strategies